In the fall of 2018 Xilinx made an interesting shift from being exclusively a “chip supplier”, specifically FPGAs, to a system level supplier. By system I mean a network interface card that could directly be used in a server. Until that time, Xilinx had relied on its end-customers or third parties to build “boards” using Xilinx FPGAs. For example, Cisco would design a hardware board around a Xilinx FPGA and then ship to its end customers. There were also companies who specialized in just taking Xilinx FPGAs, putting them on a generic board and shipping them out as “raw” cards. The term raw is used here to mean there was no pre-loaded FPGA code or intellectual property on the board. The idea is that other companies would take the raw cards, load their FPGA code on it and ship to their end customers.
With the announcement of the Alveo line of “adaptable accelerator cards”, Xilinx became not only an FPGA supplier, but also a system level supplier. This has some advantages and disadvantages. The obvious advantage is that end customers can now get both FPGAs and cards directly from one source. The disadvantage is that Xilinx runs the risk of entering some of its customers markets, most significantly the suppliers of raw boards, and in turn alienates them. The reality however is that since there are only two viable FPGA vendors in the world (Intel with its Altera acquisition being the other), there is very little risk of end-customers revolting and moving away from Xilinx.